PhD candidate · ECE · Rutgers
I build both ends of a biomedical instrument — the low-noise sensor that captures the signal, and the deep-learning model that reads it out.
Take a low-SNR measurement no one trusts yet, and turn it into a number a clinician can act on.
Machine-learning people rarely touch the hardware; sensor people rarely train a model. I work both ends. I build fluorescence and Hall-effect biosensors — the low-noise analog front-end, the COMSOL device physics, the microfluidics — then the deep-learning readout that pulls a signal out of the noise floor and turns it into a number a clinician can act on. Before Rutgers: two years in SPECT R&D at Siemens Healthineers.
I work across the full chain — low-noise analog front ends, COMSOL multiphysics models of the sensor, and the deep-learning code that reads it out — because in point-of-care diagnostics the hardware and the algorithm only work when they are designed together. Most of what I build ends up as an open, no-install web tool so collaborators and clinicians can run it without setup.
Each of these is deployed and interactive — open one and change a parameter. Each of the four also has an on-site deep-dive explainer.
FIG. 1
3D Three.js simulator with live signal trace
A real-time 3D microfluidic channel where magnetic beads flow past a Hall sensor under a Halbach array, with the Hall-voltage waveform updating live. Validated against COMSOL to under 6% error.
FIG. 2
CNN-based cell quantification
Counts leukocytes in fluorescent micrographs from a browser upload — a 12-layer CNN scores image tiles and sums them. From the IEEE Access leukocyte-quantification work.
FIG. 3
Self-supervised fluorescence image recovery
Recovers low-SNR smartphone-microscope images with no paired training data, pairing histogram thresholding with a Deep Image Prior. Published in The Analyst (RSC).
FIG. 4
Interactive multiphysics simulator
Simulates superparamagnetic-microbead detection on a back-gated CVD graphene Hall sensor — Langevin magnetization, dipole stray fields, and gate-tunable carrier density, all recomputed live.
FIG. 5
Automated neuron-health image analysis
Quantifies neuron viability, neurite length, and branching from brightfield microscopy using Frangi vesselness filtering, in an interactive dashboard for antioxidant-treatment screening.
FIG. 6
Graphene FET lab instrument
A label-free graphene field-effect biosensor for C-reactive protein: a 28-channel Graphenea mGFET with NI-DAQ readout and reference-subtracted resistance shifts (R² ≈ 0.99999).
The Analyst, Royal Society of Chemistry (2025)
Shows you can recover low-SNR fluorescence images with no paired training data — lifting PSNR from 15.6 to 27.1 dB — which makes deep-learning denoising usable in point-of-care settings where clean references don't exist.
IEEE Access (2022)
The end-to-end case that a $12 smartphone microscope plus a CNN can count white blood cells accurately enough to be clinically meaningful.
Multiphysics design study
The physics behind uFlow and GHSL: how a Halbach array and Hall sensor geometry can be co-optimized in simulation before anything is fabricated.
Peer-reviewed
Generalizes the counting approach beyond leukocytes to fluorescent particles, tying imaging-derived features to a clinical readout.
Peer-reviewed
The infrastructure side of the work at Siemens Healthineers — turning raw, semi-structured detector scans into a queryable, integrity-checked research dataset.
In progress — Graphene Hall-Effect Biosensor for Multiplexed Biomarker Detection — under review, IEEE Sensors Journal.
The fastest way to reach me is email — I read everything.
Based in New Brunswick, NJ · Open to on-site NYC, hybrid, or remote for the right team